Finally, let us discuss the biological question of what brain mechanisms generate synchronous oscillations in the visual cortex. There are two contrasting views about this issue. One of them is that a global feedback mechanism gives rise to coherent oscillations. This view is well demonstrated by the comparator model of Kammen et al. , where a global comparator receives input from every oscillator of a population of uncoupled phase oscillators (see ), and feeds back a function of the average phase of the population to every oscillator. The comparator is able to synchronize the oscillator population. Based on this result and the inability of locally coupled phase models to reach global synchrony, Kammen et al.  proposed that synchronous oscillations in the visual cortex are produced by a feedback loop from a common locus, presumably in the thalamus. Notice that the thalamus sends projections to and receives input from almost the entire cortex, and thus provides a necessary anatomical substrate. This view also finds support from Llinás and his colleagues [38,45] who, based on their experimental results, also suggested that synchronization in the brain is achieved by a mechanism similar to the comparator model. More precisely, they suggested that the thalamus play the role of synchronizing cortical oscillations. The feedback view is also expressed in Grossberg and Somers .
Another view is that synchronous oscillations are generated by lateral connections within the visual cortex. This view has been proposed from different perspectives by a number of researchers [14,24,34,51,52,57,66,73]. Singer and colleagues [14,24,51,52] have consistently advocated that coherent oscillations in the visual cortex result from lateral interactions within the cortex. On the experimental side, they reported that interhemispheric synchronization is disrupted after severing the corpus callosum , which suggests that cortico-cortical connections are critical for establishing synchronous oscillations. Another line of evidence was reported by Löwel and Singer  who found that correlated neuronal activity helps select horizontal connections during development. Our view and theoretical reasoning supporting lateral connections are given in the Why Local ... section. As summarized above, recent advances on understanding synchronous properties of oscillator networks have firmly established the theoretical legitimacy of this view.
Although structurally similar to the comparator model of Kammen et al. , the global inhibitor in Terman and Wang  serves an entirely different purpose: Desynchronization. Rapid synchrony occurs regardless of the global inhibitor. Since both the comparator and the global inhibitor are implicated to be located in the thalamus, the disputes may be settled by the following experiment. Assume that a visual scene consists two simple objects (something like Figure 6a, or even simpler). The model of Terman and Wang  predicts that the global inhibitor oscillates with a frequency double that of the oscillators on the network, because the global inhibitor is excited by each group of synchronous oscillators. The prediction implies that if the visual cortex shows 40 Hz oscillations then the thalamus oscillates with a frequency of 80 Hz. In contrast, the comparator model assumes that the feedback loop between the thalamus and the cortex produces synchrony, and thus would predict that the thalamus oscillates with the same frequency as the visual cortex. The presence of two objects is a critical condition. Otherwise both models predict the same frequency of the oscillations in the visual cortex and the thalamus.